This repo is a summary of the Permafrost Hackathon, which took place at the ETH Zurich end of November 2019.
For more background information check the following blogpost.
The presented code and approach is the result of Team Aroma. Visit the Hackathon repo for other contributions.
Our team focused on the detection of movements on stationary images. Below is an example. It might be hard to spot but there are mountaineers on the pictures. However, the pictures are four minutes apart. In column C the difference image (A - B) is depicted. Clearly, the movements of mountaineers can be spotted.
There is some noise in the difference image from alterations in light. Hence, we implemented a method (based on Turgay Celik's paper Unsupervised Change Detection in Satellite Images Using Principal Components Analysis and k-Means Clustering, 2009) to extract the relevant movements/changes in our sequential images. Check the comparison below.
Visit the above mentioned hackathon repo to download the relevant data source.
For reproducability only the data source timelapse_images_fast.zip is necessary.
Make sure your folder structure looks like this:
.
├── 00_input
│ └── image001.jpg
├── 01_data
│ └── timelapse_images_fast
│ ├── 2017-01-01
│ ├── 20170101_070009.JPG
│ ├── 20170101_070409.JPG
│ └── ...
│ ├── 2017-01-02
│ └── ...
├── 02_code
│ ├── helperfunctions.py
│ └── movement_detection.py
├── LICENSE
├── README.md
├── requirements.txt
└── setup.sh
If there are any questions, don't hesitate to shoot me an email.